When tested against leading open-source models from the CoIR leaderboard, Augment's retrieval system significantly outperformed them on real tasks—even though those models ranked higher on synthetic tests. This demonstrates a critical gap between benchmark performance and practical utility.

1 likes 1 replies